This commit introduces a shell script for data symbol testing.
The testing is designed a data structure with 64-byte alignment, it has
two fields "data1" and "data2", and other fields are reserved.
Using "perf mem" command, we can record and report memory samples for a
self-contained workload with 1 second duration. If have no any memory
sample for the data structure "buf1", it reports failure; and by
checking the offset in structure "buf1", if any memory accessing is not
for "data1" and "data2" fields, it means wrong data symbol parsing and
returns failure.
Signed-off-by: Leo Yan <[email protected]>
---
Channges from v3:
- Add specific testing chunk for AMD CPUs (Ravi).
tools/perf/tests/shell/test_data_symbol.sh | 93 ++++++++++++++++++++++
1 file changed, 93 insertions(+)
create mode 100755 tools/perf/tests/shell/test_data_symbol.sh
diff --git a/tools/perf/tests/shell/test_data_symbol.sh b/tools/perf/tests/shell/test_data_symbol.sh
new file mode 100755
index 000000000000..43ab1e37407c
--- /dev/null
+++ b/tools/perf/tests/shell/test_data_symbol.sh
@@ -0,0 +1,93 @@
+#!/bin/bash
+# Test data symbol
+
+# SPDX-License-Identifier: GPL-2.0
+# Leo Yan <[email protected]>, 2022
+
+skip_if_no_mem_event() {
+ perf mem record -e list 2>&1 | egrep -q 'available' && return 0
+ return 2
+}
+
+skip_if_no_mem_event || exit 2
+
+# skip if there's no compiler
+if ! [ -x "$(command -v cc)" ]; then
+ echo "skip: no compiler, install gcc"
+ exit 2
+fi
+
+TEST_PROGRAM=$(mktemp /tmp/__perf_test.program.XXXXX)
+PERF_DATA=$(mktemp /tmp/__perf_test.perf.data.XXXXX)
+
+check_result() {
+ # The memory report format is as below:
+ # 99.92% ... [.] buf1+0x38
+ result=$(perf mem report -i ${PERF_DATA} -s symbol_daddr -q 2>&1 |
+ awk '/buf1/ { print $4 }')
+
+ # Testing is failed if has no any sample for "buf1"
+ [ -z "$result" ] && return 1
+
+ while IFS= read -r line; do
+ # The "data1" and "data2" fields in structure "buf1" have
+ # offset "0x0" and "0x38", returns failure if detect any
+ # other offset value.
+ if [ "$line" != "buf1+0x0" ] && [ "$line" != "buf1+0x38" ]; then
+ return 1
+ fi
+ done <<< "$result"
+
+ return 0
+}
+
+cleanup_files()
+{
+ echo "Cleaning up files..."
+ rm -f ${PERF_DATA}
+ rm -f ${TEST_PROGRAM}
+}
+
+trap cleanup_files exit term int
+
+# compile test program
+echo "Compiling test program..."
+cat << EOF | cc -o ${TEST_PROGRAM} -x c -
+typedef struct _buf {
+ char data1;
+ char reserved[55];
+ char data2;
+} buf __attribute__((aligned(64)));
+
+static buf buf1;
+
+int main(void) {
+ for (;;) {
+ buf1.data1++;
+ buf1.data2 += buf1.data1;
+ }
+ return 0;
+}
+EOF
+
+echo "Recording workload..."
+
+# perf mem/c2c internally uses IBS PMU on AMD CPU which doesn't support
+# user/kernel filtering and per-process monitoring, spin program on
+# specific CPU and test in per-CPU mode.
+is_amd=$(egrep -c 'vendor_id.*AuthenticAMD' /proc/cpuinfo)
+if (($is_amd >= 1)); then
+ perf mem record -o ${PERF_DATA} -- -C 0 -- taskset -c 0 $TEST_PROGRAM &
+else
+ perf mem record --all-user -o ${PERF_DATA} -- $TEST_PROGRAM &
+fi
+
+PERFPID=$!
+
+sleep 1
+
+kill $PERFPID
+wait $PERFPID
+
+check_result
+exit $?
--
2.34.1
Hi Leo,
On Fri, Sep 30, 2022 at 5:09 AM Leo Yan <[email protected]> wrote:
>
> This commit introduces a shell script for data symbol testing.
>
> The testing is designed a data structure with 64-byte alignment, it has
> two fields "data1" and "data2", and other fields are reserved.
>
> Using "perf mem" command, we can record and report memory samples for a
> self-contained workload with 1 second duration. If have no any memory
> sample for the data structure "buf1", it reports failure; and by
> checking the offset in structure "buf1", if any memory accessing is not
> for "data1" and "data2" fields, it means wrong data symbol parsing and
> returns failure.
>
> Signed-off-by: Leo Yan <[email protected]>
> ---
>
> Channges from v3:
> - Add specific testing chunk for AMD CPUs (Ravi).
>
> tools/perf/tests/shell/test_data_symbol.sh | 93 ++++++++++++++++++++++
> 1 file changed, 93 insertions(+)
> create mode 100755 tools/perf/tests/shell/test_data_symbol.sh
>
> diff --git a/tools/perf/tests/shell/test_data_symbol.sh b/tools/perf/tests/shell/test_data_symbol.sh
> new file mode 100755
> index 000000000000..43ab1e37407c
> --- /dev/null
> +++ b/tools/perf/tests/shell/test_data_symbol.sh
> @@ -0,0 +1,93 @@
> +#!/bin/bash
> +# Test data symbol
> +
> +# SPDX-License-Identifier: GPL-2.0
> +# Leo Yan <[email protected]>, 2022
> +
> +skip_if_no_mem_event() {
> + perf mem record -e list 2>&1 | egrep -q 'available' && return 0
> + return 2
> +}
> +
> +skip_if_no_mem_event || exit 2
> +
> +# skip if there's no compiler
> +if ! [ -x "$(command -v cc)" ]; then
> + echo "skip: no compiler, install gcc"
> + exit 2
> +fi
> +
> +TEST_PROGRAM=$(mktemp /tmp/__perf_test.program.XXXXX)
> +PERF_DATA=$(mktemp /tmp/__perf_test.perf.data.XXXXX)
> +
> +check_result() {
> + # The memory report format is as below:
> + # 99.92% ... [.] buf1+0x38
> + result=$(perf mem report -i ${PERF_DATA} -s symbol_daddr -q 2>&1 |
> + awk '/buf1/ { print $4 }')
> +
> + # Testing is failed if has no any sample for "buf1"
> + [ -z "$result" ] && return 1
> +
> + while IFS= read -r line; do
> + # The "data1" and "data2" fields in structure "buf1" have
> + # offset "0x0" and "0x38", returns failure if detect any
> + # other offset value.
> + if [ "$line" != "buf1+0x0" ] && [ "$line" != "buf1+0x38" ]; then
> + return 1
> + fi
> + done <<< "$result"
> +
> + return 0
> +}
> +
> +cleanup_files()
> +{
> + echo "Cleaning up files..."
> + rm -f ${PERF_DATA}
> + rm -f ${TEST_PROGRAM}
> +}
> +
> +trap cleanup_files exit term int
> +
> +# compile test program
> +echo "Compiling test program..."
> +cat << EOF | cc -o ${TEST_PROGRAM} -x c -
> +typedef struct _buf {
> + char data1;
> + char reserved[55];
> + char data2;
> +} buf __attribute__((aligned(64)));
> +
> +static buf buf1;
> +
> +int main(void) {
> + for (;;) {
> + buf1.data1++;
> + buf1.data2 += buf1.data1;
> + }
> + return 0;
> +}
> +EOF
> +
> +echo "Recording workload..."
> +
> +# perf mem/c2c internally uses IBS PMU on AMD CPU which doesn't support
> +# user/kernel filtering and per-process monitoring, spin program on
> +# specific CPU and test in per-CPU mode.
> +is_amd=$(egrep -c 'vendor_id.*AuthenticAMD' /proc/cpuinfo)
> +if (($is_amd >= 1)); then
> + perf mem record -o ${PERF_DATA} -- -C 0 -- taskset -c 0 $TEST_PROGRAM &
I'm curious why it has "--" before "-C 0"..
Thanks,
Namhyung
> +else
> + perf mem record --all-user -o ${PERF_DATA} -- $TEST_PROGRAM &
> +fi
> +
> +PERFPID=$!
> +
> +sleep 1
> +
> +kill $PERFPID
> +wait $PERFPID
> +
> +check_result
> +exit $?
> --
> 2.34.1
>
>> +# perf mem/c2c internally uses IBS PMU on AMD CPU which doesn't support
>> +# user/kernel filtering and per-process monitoring, spin program on
>> +# specific CPU and test in per-CPU mode.
>> +is_amd=$(egrep -c 'vendor_id.*AuthenticAMD' /proc/cpuinfo)
>> +if (($is_amd >= 1)); then
>> + perf mem record -o ${PERF_DATA} -- -C 0 -- taskset -c 0 $TEST_PROGRAM &
>
> I'm curious why it has "--" before "-C 0"..
IIRC, any command line options placed after -- will be carry forwarded
as is to perf record.
perf mem record [options] does not support -C option. Which means -C 0
will be ignored if user run `perf mem record -C 0`. Ex:
$ sudo ./perf --debug perf-event-open mem record -C 0
[...]
sys_perf_event_open: pid -1 cpu 0 group_fd -1 flags 0 = 5
sys_perf_event_open: pid -1 cpu 1 group_fd -1 flags 0 = 6
sys_perf_event_open: pid -1 cpu 2 group_fd -1 flags 0 = 7
sys_perf_event_open: pid -1 cpu 3 group_fd -1 flags 0 = 8
OTOH, if I place -C 0 after -- then perf opens event only on cpu0:
$ sudo ./perf --debug perf-event-open mem record -- -C 0
[...]
sys_perf_event_open: pid -1 cpu 0 group_fd -1 flags 0 = 7
Now, perf mem [options] supports -C options but I don't know why
it's not working for me:
$ sudo ./perf --debug perf-event-open mem -C 0 record
[...]
sys_perf_event_open: pid -1 cpu 0 group_fd -1 flags 0 = 5
sys_perf_event_open: pid -1 cpu 1 group_fd -1 flags 0 = 6
sys_perf_event_open: pid -1 cpu 2 group_fd -1 flags 0 = 7
sys_perf_event_open: pid -1 cpu 3 group_fd -1 flags 0 = 8
Thanks,
Ravi
On Fri, Sep 30, 2022 at 10:13 PM Ravi Bangoria <[email protected]> wrote:
>
> >> +# perf mem/c2c internally uses IBS PMU on AMD CPU which doesn't support
> >> +# user/kernel filtering and per-process monitoring, spin program on
> >> +# specific CPU and test in per-CPU mode.
> >> +is_amd=$(egrep -c 'vendor_id.*AuthenticAMD' /proc/cpuinfo)
> >> +if (($is_amd >= 1)); then
> >> + perf mem record -o ${PERF_DATA} -- -C 0 -- taskset -c 0 $TEST_PROGRAM &
> >
> > I'm curious why it has "--" before "-C 0"..
>
> IIRC, any command line options placed after -- will be carry forwarded
> as is to perf record.
>
> perf mem record [options] does not support -C option. Which means -C 0
> will be ignored if user run `perf mem record -C 0`. Ex:
>
> $ sudo ./perf --debug perf-event-open mem record -C 0
> [...]
> sys_perf_event_open: pid -1 cpu 0 group_fd -1 flags 0 = 5
> sys_perf_event_open: pid -1 cpu 1 group_fd -1 flags 0 = 6
> sys_perf_event_open: pid -1 cpu 2 group_fd -1 flags 0 = 7
> sys_perf_event_open: pid -1 cpu 3 group_fd -1 flags 0 = 8
>
> OTOH, if I place -C 0 after -- then perf opens event only on cpu0:
>
> $ sudo ./perf --debug perf-event-open mem record -- -C 0
> [...]
> sys_perf_event_open: pid -1 cpu 0 group_fd -1 flags 0 = 7
>
> Now, perf mem [options] supports -C options but I don't know why
> it's not working for me:
Thanks for checking this!
It seems -C works only for perf mem report and eats it for record.
I think we need to fix the option handling.
Thanks,
Namhyung
>
> $ sudo ./perf --debug perf-event-open mem -C 0 record
> [...]
> sys_perf_event_open: pid -1 cpu 0 group_fd -1 flags 0 = 5
> sys_perf_event_open: pid -1 cpu 1 group_fd -1 flags 0 = 6
> sys_perf_event_open: pid -1 cpu 2 group_fd -1 flags 0 = 7
> sys_perf_event_open: pid -1 cpu 3 group_fd -1 flags 0 = 8
>
> Thanks,
> Ravi
On Tue, Oct 04, 2022 at 11:34:45AM -0700, Namhyung Kim wrote:
> On Fri, Sep 30, 2022 at 10:13 PM Ravi Bangoria <[email protected]> wrote:
> >
> > >> +# perf mem/c2c internally uses IBS PMU on AMD CPU which doesn't support
> > >> +# user/kernel filtering and per-process monitoring, spin program on
> > >> +# specific CPU and test in per-CPU mode.
> > >> +is_amd=$(egrep -c 'vendor_id.*AuthenticAMD' /proc/cpuinfo)
> > >> +if (($is_amd >= 1)); then
> > >> + perf mem record -o ${PERF_DATA} -- -C 0 -- taskset -c 0 $TEST_PROGRAM &
> > >
> > > I'm curious why it has "--" before "-C 0"..
> >
> > IIRC, any command line options placed after -- will be carry forwarded
> > as is to perf record.
> >
> > perf mem record [options] does not support -C option. Which means -C 0
> > will be ignored if user run `perf mem record -C 0`. Ex:
> >
> > $ sudo ./perf --debug perf-event-open mem record -C 0
> > [...]
> > sys_perf_event_open: pid -1 cpu 0 group_fd -1 flags 0 = 5
> > sys_perf_event_open: pid -1 cpu 1 group_fd -1 flags 0 = 6
> > sys_perf_event_open: pid -1 cpu 2 group_fd -1 flags 0 = 7
> > sys_perf_event_open: pid -1 cpu 3 group_fd -1 flags 0 = 8
> >
> > OTOH, if I place -C 0 after -- then perf opens event only on cpu0:
> >
> > $ sudo ./perf --debug perf-event-open mem record -- -C 0
> > [...]
> > sys_perf_event_open: pid -1 cpu 0 group_fd -1 flags 0 = 7
> >
> > Now, perf mem [options] supports -C options but I don't know why
> > it's not working for me:
>
> Thanks for checking this!
>
> It seems -C works only for perf mem report and eats it for record.
> I think we need to fix the option handling.
Thanks for digging into the issue and fixing, Ravi & Namhyung.
I will send a new patch to drop "--" for CPU list.
Thanks,
Leo